Japanese Review Classifier — sentiment & complaints

For sellers, brands and AI agents that have Japanese reviews — from a Rakuten (楽天) review scraper, any other Apify dataset, or pasted as text — and need every review labelled with complaint type, sentiment and purchase motive, for $0.50 per 1,000 reviews, with no scraping, prompt writing or LLM key.

• Complaint types — delivery, quality/defect, size/fit, price/value, shop response, packaging, no effect, skin/body reaction, other (multi-label, each with a probability)
• Sentiment — positive / neutral / negative
• Purchase motive — price/discount, reviews & reputation, brand, gift, repurchase, unknown
• Your own labels — up to 10 yes/no criteria in plain English or Japanese (e.g. "mentions charging speed", "配達員の対応が悪い")

Labels come with English keys and Japanese names (e.g. quality_defect / 品質・不良, field labelJa).

Use it to: find why Rakuten buyers leave 1-star reviews (楽天レビュー分析) · compare complaint mix against competitors · separate shipping and packaging problems from product defects · tag Japanese reviews for a dashboard · Japanese review sentiment analysis (レビュー分析) at scale.

Output sample

Real rows from run GrEqVhVam3XUpbStj (2026-10-08), reviews of power banks on Rakuten, with one custom label ("mentions charging speed"). English translations are added here for readers; the Actor returns the original text.

text (Japanese)	English (added)	complaint (probability)	sentiment	motive	custom: mentions charging speed
早くに発送してくださり ありがとうございます。 予想より重たかったです。 色はかわいいです。 本体の充電 時間がかかりすぎて そこが難点…	Shipped fast, thanks. Heavier than expected. Cute colour. Charging the unit takes far too long — that's the downside.	other (0.70)	neutral (0.73)	unknown	true
すぐに届きましたが、バッテリーを充電しても 39%から永遠に上がりません。 不良品ですかね。 困ります。交換してもらいたいです。	Arrived quickly, but the battery never charges past 39%. Defective? I want an exchange.	quality_defect (0.97)	negative (1.00)	unknown	false
箱が潰れて中身もどうなってるかわかりませんて言われました どういう扱いしてるんですかね？ 新しく商品を送り返してくれましたが、再発送の…	Told the box was crushed and the contents might be damaged… resent without notice, no apology. Never again.	packaging (0.96), customer_service (0.91), delivery (0.60)	negative (1.00)	unknown	false
安定の商品でした。有難うございます。 リピートの際には、また購入させていただきます。	Reliable product, thank you. I'll buy again.	none (0.98)	positive (0.99)	unknown	false

Each row also has labelJa names, the rating and the ID fields you choose (reviewId, productId, …), and in full mode complaintScores for every complaint type.

Input example

The form default — two pasted reviews, no dataset needed (about $0.001, 2 seconds):

{
  "texts": [
    "ダンボールが潰れて届きました。中身は無事でしたが、梱包をもう少し丁寧にしてほしいです。",
    "すぐ壊れました。充電できません。返品したいです。"
  ]
}

To classify a Rakuten review scraper's output, pass its dataset instead — the text, rating and ID fields are detected automatically (checked on run b2QEJvgEe4N9TURc1: 15 reviews with reviewId, productId and rating kept):

{
  "datasetId": "YOUR_RAKUTEN_REVIEW_DATASET_ID",
  "customLabels": ["mentions charging speed"]
}

Pricing

Pay only for classified reviews — no subscription.

Event	Price	When
review-judged	$0.0005	One review classified (complaint types, sentiment, purchase motive and any custom labels).

That is $0.50 per 1,000 reviews. First run with the form defaults: about $0.001 (2 reviews, 2 seconds).

Cost examples

Reviews	Cost
100	$0.05
1,000	$0.50
10,000	$5.00
100,000	$50.00

With the free $5 monthly Apify credit you can classify about 10,000 reviews.

Items without review text are skipped and not charged. Reviews that fail after retries are reported with an error field and not charged. If you set a maximum cost per run, the Actor stops cleanly when it is reached.

Works with

Rakuten review scrapers on Apify Store — run one, then pass its dataset to this Actor. Field detection was checked against real output of both.

Source	Scraper on Apify Store	Text field	Rating	IDs kept
Rakuten reviews	Rakuten Japan Reviews Scraper (piotrv1001)	text	rating	reviewId, productId
Rakuten reviews	Rakuten Ichiba Reviews Scraper (axlymxp)	body	rating	item_id, shop_id

Also: the Apify API and JavaScript/Python clients · Apify Schedules · Claude, Cursor and Claude Code through the Apify MCP server (next section) · for Korean reviews, our Korean Review Classifier uses the same labels.

Use with Claude, Cursor or Claude Code (MCP)

Add the Apify MCP server with this Actor as a tool and ask your agent in plain language — for example "Classify these Japanese reviews and tell me the top complaint types: …" or "Classify dataset abc123 from my review scraper run and summarize the complaints." The agent calls the tool leoworks--japanese-review-classifier and reads the labels with get-dataset-items.

Claude Desktop or Cursor (mcp.json):

{
  "mcpServers": {
    "apify": {
      "url": "https://mcp.apify.com?tools=leoworks/japanese-review-classifier",
      "headers": { "Authorization": "Bearer YOUR_APIFY_TOKEN" }
    }
  }
}

Claude Code: claude mcp add --transport http apify "https://mcp.apify.com?tools=leoworks/japanese-review-classifier" --header "Authorization: Bearer YOUR_APIFY_TOKEN". Leave out the header to sign in with OAuth in the browser instead. Your Apify token is in Console → Settings → API & Integrations. We verified this setup with the Apify MCP server (v0.17.3) on 2026-10-08: the agent classified a pasted review in 2.5 seconds (run GUnh4xa8RtCXm76yZ).

Output (one row per review)

{
  "index": 1,
  "text": "箱が潰れて中身もどうなってるかわかりませんて言われました\nどういう扱いしてるんですかね？\n\n新しく商品を送り返してくれましたが、再発送の連絡もなくいつのまにか届いてました\n謝罪の言葉もこちらから言うまでなく2度と買いません",
  "labels": {
    "complaint": [
      {
        "label": "packaging",
        "labelJa": "梱包",
        "probability": 0.96
      },
      {
        "label": "customer_service",
        "labelJa": "ショップ対応",
        "probability": 0.91
      },
      {
        "label": "delivery",
        "labelJa": "配送",
        "probability": 0.6
      }
    ],
    "sentiment": {
      "label": "negative",
      "labelJa": "否定",
      "probability": 1,
      "confidence": 1
    },
    "motive": {
      "label": "unknown",
      "labelJa": "不明",
      "probability": 0.9,
      "confidence": 0.88
    }
  }
}

When no complaint type passes the threshold, complaint is [{ "label": "none", "labelJa": "不満なし" }]. Minimal mode returns label keys only.

Summary by product (REPORT)

Each run also saves a REPORT record (Output tab → Summary by product) at no extra charge: for every product, the complaint rate and complaint mix, sentiment shares, purchase motives, average rating and the 3 strongest complaint reviews — plus the same for all reviews together. Products are grouped by summaryGroupField (detected automatically from fields such as productId, productName or placeId when empty).

{
  "groupField": "productId",
  "groups": [{
    "group": "A",
    "reviews": 3,
    "averageRating": 2.67,
    "complaintRate": 0.667,
    "complaints": [{ "label": "delivery", "count": 1, "share": 0.333 }, { "label": "quality_defect", "count": 1, "share": 0.333 }],
    "sentiment": { "negative": 0.667, "positive": 0.333 },
    "motive": { "unknown": 0.667, "price": 0.333 },
    "exampleComplaints": [{ "complaint": "delivery", "rating": 2, "text": "配送が1週間もかかりました。遅すぎます" }]
  }]
}

Accuracy

Measured on hand-labelled Japanese Rakuten reviews (2026-10-07/08). The questions were adjusted on the first set (a crushed outer box counts as packaging, not a product defect), then checked on a new set of different products.

Set	Products	Reviews (1–2 stars)	Complaint type	Sentiment
New check set — not used for adjusting	Power banks	47 (32)	91%	96%
First set	Bottled water	50 (30)	96%	100%
Second set	Mugs, power banks	40 (5)	100%	100%

Probabilities are calibrated — raise Complaint threshold for fewer, surer labels. Automated labels can be wrong; check samples before making big decisions.

Limits

Item	Limit
Review length	First 4,000 characters are used
Custom labels	Up to 10, each up to 200 characters
Dataset size	Any — datasets are read in pages of 1,000
Language	Japanese (measured). Other languages: see our Korean and Spanish classifiers
Speed	About 100 reviews in 5 seconds
Data	Only the text, rating and the ID fields you choose are sent for classification; reviewer names are not output

FAQ

Which AI makes the judgments? Jev, TypeSafe's decision model (version jev-1.13.0, pinned). Jev answers each label with a calibrated probability instead of generated text, so the same input gets the same answer from run to run. Only the review text, its rating and your custom labels are sent to Jev; reviewer names and other fields are not.

Does it scrape Rakuten? No. It classifies reviews you already have — run a Rakuten review scraper first (see Works with) or paste texts.

Does it generate text or summaries? No. It only assigns labels with probabilities — fast, cheap and consistent.

Disclaimer

Independent tool — not affiliated with, endorsed by or sponsored by Rakuten Group, or by the authors of the scrapers listed above. Names are used only to describe compatible data sources.

Reviews and support

If this Actor saved you time, a short review on Apify Store helps others find it. Questions or a dataset whose fields are not detected? Open an issue in the Issues tab — we answer within a day.

Changelog

See the Changelog tab.